Third party funded individual grant
Start date : 01.04.2021
End date : 31.03.2023
Noise in the brain is often seen as disruptive, but it can actually help neurons process information by creating resonance effects. These effects, observed in both single neurons and networks, depend on factors like neuron properties, network structure, and the type of noise. Past research has mostly studied these mechanisms separately and in fixed (non-adaptive) networks.
This project aims to go further by exploring how different resonance mechanisms—coherence resonance (CR), self-induced stochastic resonance (SISR), and recurrence resonance (RR)—can be optimized in adaptive (co-evolving) neural networks with different structures (such as small-world, random, or multilayer networks). The study will use realistic Hodgkin-Huxley neuron models that evolve through learning rules based on timing and activity.
By identifying the best conditions and strategies for these resonance effects, the project seeks to uncover new principles of how the brain optimizes information processing.
The functional role of noise is a long-standing question in neurobiology. Although noise is often viewed as undesirable, its resonance effects are well established and play a crucial role in how neurons encode information. Resonance has been observed in both individual neurons and networks, and different types of noise-induced resonance mechanisms arise under different conditions, depending on neuron parameters, synaptic connections, network topology, and noise sources. Previous research has largely focused on optimizing these mechanisms in non-adaptive neural networks and treating them independently of one another. What remains missing is a comprehensive understanding of how information processing can be optimized through co-evolutionary (adaptive) neural networks and by exploiting the interplay between two or more resonance mechanisms.
The main objective of this project is to design and classify, in terms of efficiency, optimization schemes for three noise-induced resonance mechanisms in co-evolutionary biological neural networks: coherence resonance (CR), self-induced stochastic resonance (SISR), and recurrence resonance (RR). The study will examine these mechanisms in different network structures, including motifs, scale-free networks, small-world networks, random networks, and multilayer networks. These networks will be based on the biophysical Hodgkin-Huxley (HH) neuron model and will evolve according to either the spike-timing-dependent plasticity learning rule or a temporal activity-dependent structural plasticity rule. Before moving to HH networks, the necessary conditions for CR and SISR in an isolated HH neuron will be determined using geometric singular perturbation theory and stochastic multi-dimensional reaction rate theory.
Depending on the network topology, synaptic properties, and learning rules, different optimization schemes for CR, SISR, and RR will be designed and assessed for efficiency. By addressing these issues in adaptive neural networks, this project promises to push the understanding of optimal neural coding and information processing to new frontiers.